EOG vs XLE: next 20-session relative return after Brent up / SPY down
Oil up, stocks down, and the high-beta driller that's supposed to outrun the diversified energy ETF: that was the setup tested across 78 days since late 2023. EOG did beat XLE over the following 20 sessions, but only by 0.38 percentage points on average, and it trailed on 51% of those signals.
The average is propped up by a handful of violent stretches — top spreads near +11% — while the typical outcome went the other way. On all other days the same EOG-minus-XLE spread averaged -0.70%, so the regime does flip the relationship by about 1.1 points. The direction matches the thesis; the reliability doesn't.
The paired gap inside the signal is a statistical coin flip, and EOG already lives inside XLE, so this was never a clean long-short. The full breakdown is below.
Over the past ~3 years, when Brent crude's 20-day return is positive while SPY's 20-day return is negative, does EOG outperform XLE over the next 20 trading sessions? I expect high-beta upstream torque to lead diversified energy equities when oil rallies into a risk-off tape, as the supply-driven crude bid gets repriced most directly into EOG.
How this was measured
Minute bars for EOG, XLE, and SPY were resampled to daily closes and joined on common trading dates. Daily Brent was shifted one business day to approximate T+1 publication and then forward-filled onto the price calendar. A signal day is one where Brent's trailing 20-session return is positive and SPY's trailing 20-session return is negative. For each day, the next-20-session returns of EOG and XLE are measured close-to-close, and EOG-minus-XLE is the relative outcome. Signal-day results are compared with all other days and tested via a paired EOG-vs-XLE t-test and a Welch test on the spread.
The key numbers
Reading the numbers
On the 78 days when Brent was up and SPY was down over the prior 20 sessions, EOG averaged +0.94% over the next 20 sessions versus XLE's +0.57%, a +0.38% edge. The edge versus non-signal days is bigger at +1.08%, but the paired EOG-vs-XLE gap is not statistically clear (p=0.44).
The charts
Each point is a signal day, plotting how strong Brent's 20-day move was against how much EOG subsequently beat or lagged XLE over the next 20 sessions. Brent moves ranged from about +0.3% to +70.3%, averaging +17.6%, while the EOG-minus-XLE spread ranged from -8.86% to +11.00%, averaging +0.38%. The scatter does not show a simple pattern where bigger crude rallies line up with bigger EOG outperformance, which tempers the idea that the crude bid reprices into EOG in a straight-line way.
Next-20d summary by regime
| Sample | N | Mean EOG fwd20 | Mean XLE fwd20 | Mean EOG-XLE | Win rate |
|---|---|---|---|---|---|
| Signal: Brent up / SPY down | 78 | 0.0094 | 0.0057 | 0.0038 | 0.4872 |
| Other days | 589 | 0.0096 | 0.0166 | -0.007 | 0.4363 |
| All days | 667 | 0.0096 | 0.0153 | -0.0057 | 0.4423 |
Signal-day detail (sorted by EOG minus XLE)
| Date | Brent 20d | SPY 20d | EOG fwd20 | XLE fwd20 | EOG-XLE |
|---|---|---|---|---|---|
| 2026-02-10 | 0.0934 | -0.0003 | 0.1737 | 0.0637 | 0.11 |
| 2026-02-26 | 0.0058 | -0.012 | 0.2263 | 0.1265 | 0.0998 |
| 2026-02-04 | 0.1113 | -0.0061 | 0.1573 | 0.0677 | 0.0896 |
| 2026-02-05 | 0.1457 | -0.0235 | 0.1661 | 0.0852 | 0.0809 |
| 2026-02-27 | 0.0107 | -0.0129 | 0.2055 | 0.1253 | 0.0802 |
| 2026-02-11 | 0.0858 | -0.0017 | 0.1289 | 0.049 | 0.0799 |
| 2026-02-09 | 0.1123 | -0.0011 | 0.1153 | 0.0359 | 0.0793 |
| 2026-03-02 | 0.0045 | -0.0089 | 0.1601 | 0.0957 | 0.0643 |
| 2026-03-04 | 0.2298 | -0.0026 | 0.1155 | 0.052 | 0.0635 |
| 2026-02-12 | 0.0583 | -0.0128 | 0.1335 | 0.07 | 0.0635 |
| 2026-02-13 | 0.0135 | -0.0173 | 0.1261 | 0.0633 | 0.0629 |
| 2026-02-25 | 0.0518 | -0.0072 | 0.1719 | 0.112 | 0.0599 |
| 2026-02-23 | 0.1114 | -0.0069 | 0.1408 | 0.0863 | 0.0546 |
| 2026-02-19 | 0.0606 | -0.0022 | 0.1287 | 0.076 | 0.0527 |
| 2026-02-17 | 0.0703 | -0.0139 | 0.1354 | 0.0903 | 0.0451 |
| 2025-01-16 | 0.1336 | -0.0219 | -0.0615 | -0.0313 | -0.0302 |
| 2024-04-24 | 0.0397 | -0.0351 | -0.0824 | -0.0507 | -0.0318 |
| 2024-04-25 | 0.0489 | -0.0257 | -0.0826 | -0.0505 | -0.0321 |
| 2025-11-20 | 0.0241 | -0.0266 | -0.0379 | -0.0055 | -0.0324 |
| 2025-04-04 | 0.0205 | -0.118 | -0.0014 | 0.0367 | -0.0381 |
| 2024-04-29 | 0.047 | -0.0241 | -0.1028 | -0.0583 | -0.0445 |
| 2025-11-17 | 0.0363 | -0.0096 | -0.071 | -0.0241 | -0.0469 |
| 2026-07-20 | 0.0583 | -0.0064 | 0.0317 | 0.0799 | -0.0482 |
| 2024-04-26 | 0.0223 | -0.0212 | -0.1004 | -0.0514 | -0.049 |
| 2024-04-30 | 0.0412 | -0.0414 | -0.0536 | -0.0037 | -0.0498 |
| 2026-07-17 | 0.0091 | -0.0017 | 0.0131 | 0.0715 | -0.0584 |
| 2026-07-30 | 0.305 | -0.0022 | -0.0068 | 0.0533 | -0.0601 |
| 2024-05-01 | 0.0068 | -0.0352 | -0.0844 | -0.0141 | -0.0703 |
| 2026-07-29 | 0.1944 | -0.0186 | -0.0071 | 0.0655 | -0.0727 |
| 2026-07-31 | 0.3274 | -0.002 | -0.0454 | 0.0432 | -0.0886 |
The takeaway
Directionally yes, but far too weak to lean on: across the 78 days since late 2023 when Brent's 20-day return was positive and SPY's was negative, EOG beat XLE by an average of 0.38 percentage points over the next 20 sessions (+0.94% vs +0.57%). That same EOG-minus-XLE spread averaged -0.70% on all other days, so the regime does flip the relationship by roughly 1.1 points — the sign is what you predicted. The catch is underneath the average: EOG only won 49% of the time, so it actually lagged in more than half of these signals, and a positive mean with a sub-50% win rate means a few blowout stretches are doing the work (top signal days show spreads of +11%, +10%, and +9%). Within the regime, the EOG-vs-XLE gap is statistically indistinguishable from noise (p = 0.44 — essentially a coin flip), and the more flattering signal-days-vs-other-days comparison (p = 0.037) is boosted by overlapping 20-day windows that inflate the effective sample size. Read it as a plausible tilt rather than an edge: the story about high-beta crude torque is not contradicted, but it is not confirmed either, and EOG sitting inside XLE means this was never a clean long-short to begin with.
The fine print
- Only 78 signal days in the window — thin evidence, and the positive average rests on a handful of outsized winners.
- Forward 20-session windows overlap across nearby signal days, so the p-values overstate how much independent information there really is.
- EOG is a component of XLE, so EOG-vs-XLE is not an independent spread and mechanically muted relative to a true long-short.
- No costs, slippage, or financing are modeled — these are return observations, not tradable P&L.